Retention science
What the Research Says About Keeping People, and How Anchor Applies It
Decades of retention research keep pointing the same direction. It is individual, it runs through the direct manager, and it depends on data people feel safe giving. Here is the science, and how Anchor turns it into something a supervisor can use.
June 14, 2026 · 6 min read
In heavy operations, energy services, maritime, clinical care, advanced manufacturing, losing the wrong person is one of the most expensive things that happens all year. Gallup puts the cost of replacing an employee at one-half to two times their annual salary, and calls that conservative. For the senior, certified, hard-to-replace people, the Center for American Progress found the figure can reach roughly 213 percent of salary.
Given that price, you would expect retention software to be sharp and specific. Most of it is neither. It assumes a workforce tethered to desks, laptops, and chat apps, and it treats engagement as a single organizational average to be raised a few points a year. The research on why people actually leave has moved well past that. Anchor was built to apply what the research has been saying for years, on the floor, where the work happens.
Retention is individual, not an average
The most consistent finding in the retention literature is also the most inconvenient for software vendors: one-size-fits-all programs underperform because they ignore individual differences. A raise, a generic perk, a company-wide initiative, these treat a hundred different people as one. The work on person-environment fit, how well a specific person matches a specific role and organization, shows that the match is personal. What holds one person holds nobody else for the same reasons.
This is why the annual survey, scored as a department average, misses so much. An average can look healthy while it hides the one person about to leave. The research points instead toward individualized intervention, aligning the work with the person in front of you.
Anchor is built around exactly that unit. One employee, one analysis, one plan. It reads people one at a time, not as a heat map, and writes a separate read for each. That is the academic call for individualized intervention turned into a working method.
The relationship that decides it is with the direct manager
Decades of work on the leader-member relationship, the specific bond between a person and their direct supervisor, identify it as one of the strongest buffers against turnover. When that relationship is strong, people stay through hard stretches. When it frays, they start to look. Gallup found that 42 percent of people who quit said their manager or organization could have done something to keep them, and that nearly half had no leader proactively discuss their satisfaction or future in their final three months.
The literature recommends targeted coaching for the manager and employee as a pair, because that is the level where the friction lives and where it can be resolved. Most software does the opposite. It routes the signal to corporate HR and away from the one person who could act on it.
Anchor puts the plan directly in the supervisor’s hand. Its engine, Pneuma, delivers the near-term flight risk, six conversation scripts, and the one move that matters for the quarter. That is the relationship-level intervention the research asks for, made specific enough to use in a real conversation.
Personality is the missing diagnostic
Engagement is a state. It rises and falls with workload, a new boss, a rough quarter. Personality is a trait, stable over years, and it shapes how a person experiences the very same job. The research increasingly pairs a standard engagement read with a measure of the Five-Factor Model of personality, often through an instrument like the IPIP, to tell two very different problems apart.
The reason matters. Low engagement can be a structural deficit, the job genuinely is the problem, or a psychological mismatch between the person and the work. The fix is different in each case, and you cannot tell which you are looking at from an engagement score alone. A highly conscientious person grinding through chronic understaffing needs something different from a person whose role has simply stopped growing.
Anchor folds this in at onboarding with a one-time personality assessment, then combines it with ongoing input from both the employee and the manager. The engine writes the analysis against all three, which is how it can tell a structural problem from a fit problem and recommend the move that actually addresses the cause.
Honest data depends on confidentiality
There is a quiet statistical trap in most engagement data. The people most likely to leave are often the least likely to answer honestly, or to answer at all, when they believe the company is watching. The dashboard fills up with the views of the people who feel safe, and the most important signal, from the person already halfway out the door, is the one that never arrives.
The way out is confidentiality that the employee can actually trust. The research is clear that psychological safety is what produces honest answers, and the direction of regulation, from data-protection law to the newer rules on workplace AI, is that a human, not an algorithm, must make consequential decisions about a person’s career.
Anchor is engineered around both. Raw answers and personality scores stay with the engine and reach no one, not HR, not corporate, not even the direct supervisor. The supervisor receives only the finished plan, and that plan is never used to grade the manager or surfaced upward. The engine analyzes and drafts, but it makes no employment decision. The intervention is a human conversation between two people who work together. That is what lets people answer honestly, which is what makes the data worth having.
None of it matters unless someone acts in time
Turnover is a process, not an event. It runs from a first quiet withdrawal to a resolved decision to a notice on the desk, and the window to change the outcome closes well before the last step. The premier upstream tool in the research, the stay conversation, works precisely because it surfaces concerns while they are still fixable.
The practical failure is timing. Most organizations are passive observers of lagging numbers, and the data they do collect is routed through approvals while the person quietly checks out. That delay between a warning sign and a real conversation is where good people are lost. It is worst in the deskless world, the void where the software stops, because field crews do not generate the digital exhaust those tools depend on.
Anchor was built for that void by a manager who spent twenty years in heavy operations. It reaches people through a simple private link, no app and no portal, and it hands the supervisor a plan in plain language before the notice lands. The science says retention is individual, manager-led, personality-informed, and confidential. Anchor is what that looks like when you have to act on it by Tuesday.
The theory has been settled for a while. The hard part was always making it usable by the person who can actually keep someone. That is the whole job Anchor set out to do.